Visual research draft on latent quorum sensing, deterministic local fragment reassembly, SHA-256 integrity checks, and graceful degradation in a SmolLM2-135M micro-LLM experiment.
An exploratory experiment examining why large language models often appear weak at chess. The hypothesis tested is that many failures are not purely strategic, but arise from unstable board representation and limited internal state tracking — effectively forcing the model to play "with its eyes closed".
A fast-mode LLM ("Aura") played against Stockfish while narrating its reasoning. The experiment observed how long coherent positional play could be maintained before board-state drift and hallucination began to dominate the game.
A spare slot for theory notes, experiments, philosophical drafts, or any visual HTML document you want to expose from the papers index.